Practical Significance of Effects from Growth Modeling of Alcohol Use Data

点击翻译按钮获取中文摘要

Abstract Longitudinal research on alcohol consumption, alcohol use disorders, and heavy episodic drinking often uses latent growth modeling analysis (GMA) and growth mixture modeling (GMM) to examine pathways to alcohol and drug abuse (including the risk and protective factors that predict them) or health consequences from abuse. Such studies of substance abuse trajectories are useful for understanding the etiology of alcohol and drug problems, and for informing efforts for both prevention and treatment. Examples of such prospective studies, both of which target adolescents, include NIAAA's National Consortium on Alcohol and Neurodevelopment in Adolescence (NCANDA) Study and NIDA's Adolescent Brain Cognitive Development (ABCD) Study. Both of these studies are also examples of big data—a current focus of NIH (e.g., their BD2K initiative)—in that they use hundreds or thousands of participants and are ideally suited for analyses with the GMA approach (which requires large samples). However, in spite of the widely recognized need for effect sizes and their confidence intervals (CIs) in statistical analyses, development of such statistics for findings from GMA and GMM has been limited. In our prior work, we have proposed a regression framework for effect size assessments and formulated an equation for a standardized effect size for GMA that transforms the trajectory difference into a standardized mean difference in the metric of Cohen's d, which is now widely used in the literature on randomized clinical trials (RCTs). We have also developed formulas for estimating the CI for the GMA effect size. We will first use that model to develop new statistics that fill critical gaps in our published work regarding GMA effect sizes and related statistics (e.g., SEs and CIs) for RCTs, which will provide a foundation for subsequent aims. Next, we will introduce new kinds of effect sizes for direct and indirect effects that would be useful to both GMA and non-GMA researchers examining mediation, although our secondary analyses that illustrate our methods will involve mediation in GMA. The proposed work will formulate equations for new effect sizes for GMA and GMM in different metrics, and Monte Carlo studies will be conducted to examine biases (and determine necessary sample sizes) for point estimates and CIs for the effect sizes for different types of GMA and GMM hypothesis tests commonly found in the literature. In addition, we will conduct a secondary analysis study of alcohol use data from both the current and previous National Longitudinal Surveys of Youth to examine whether, as predicted, the effects of gender and on growth, persistence, and desistance of alcohol use from the teen years into early adulthood has decreased between cohorts born 20 years apart—and to attempt to identify mediators responsible for the expected changes in effect sizes and whether mediation is moderated by other factors (e.g., ethnicity). Thus, this project would also illustrate the use of our new statistics in meta-analysis. Most important, we will publish tutorial articles to widely disseminate out derived equations for communicating the practical significance of findings from GMA and GMM.